Coded Parity Packet Transmission Method for Two Group Resource Allocation

Transcription

1 Coded Parity Packet Transmission Method for Two Group Resource Allocation By Hadhrami Ab. Ghani Supervised by: Dr. M.K. Gurcan Intelligent Systems and Networks Research Group A thesis submitted for a PhD degree of Imperial College London Department of Electrical and Electronic Engineering Imperial College London September 15, 2011

2 Abstract Gap value control is investigated when the number of source and parity packets is adjusted in a concatenated coding scheme whilst keeping the overall coding rate fixed. Packet-based outer codes which are generated from bit-wise XOR combinations of the source packets are used to adjust the number of both source packets. Having the source packets, the number of parity packets, which are the bit-wise XOR combinations of the source packets can be adjusted such that the gap value, which measures the gap between the theoretical and the required signal-to-noise ratio (SNR), is controlled without changing the actual coding rate. Consequently, the required SNR reduces, yielding a lower required energy to realize the transmission data rate. Integrating this coding technique with a two-group resource allocation scheme renders efficient utilization of the total energy to further improve the data rates. With a relatively small-sized set of discrete data rates, the system throughput achieved by the proposed two-group loading scheme is observed to be approximately equal to that of the existing loading scheme, which is operated with a much larger set of discrete data rates. The gain obtained by the proposed scheme over the existing equal rate and equal energy loading scheme is approximately 5 db. Furthermore, a successive interference cancellation scheme is also integrated with this coding technique, which can be used to decode and provide consecutive symbols for inter-symbol interference (ISI) and multiple access interference (MAI) mitigation. With this integrated scheme, the computational complexity is significantly reduced by eliminating matrix inversions. In the same manner, the proposed coding scheme is also incorporated into a novel fixed energy loading, which distributes packets over parallel channels, to control the gap value of the data rates although the 2

3 SNR of each code channel varies from each other. 3

4 Acknowledgments First and foremost, I would like to express my utmost gratitude to my Creator who gave me the chance to complete my thesis. Deep in my heart, I wish to acknowledge and thank Dr. M.K. Gurcan for his continuous guidance and patience with fruitful ideas in helping me to complete this project up to this stage. I am also very grateful to be accompanied with my beloved wife and little son who are so patient supporting myself morally. Throughout my study duration, Yayasan Telekom Malaysia Berhad (YTM) has been continuously supporting myself and family with a full scholarship, without which it is near impossible for me to complete my PhD. The support received from YTM, financially or morally, is very much appreciated. To all my friends and colleagues, your kind help will always be remembered and may I wish you a meaningful and peaceful life. I hereby confirm that the content of this thesis is original and with that I am ready to hold full responsibility for all results produced to complete this thesis. All related work mentioned and discussed in this thesis, to the best of my knowledge, have been properly cited. 4

5 List of Publications The content of this thesis is mainly written based on the following publications and submitted manuscripts Published Journal Paper M. Gurcan, H. Ab Ghani, J. Zhou, and A. Chungtragarn, Bit Energy Consumption Minimization for Multi-path Routing in Ad Hoc Networks, The Computer Journal, Submitted Manuscripts for Journal Publication M. K. Gurcan and Hadhrami Ab. Ghani, Coded Parity Packet Approach for Small-sized Packet Transmission in Wireless Ad-hoc Networks, submitted to European Transactions on Communications, Dec M. K. Gurcan and Hadhrami Ab. Ghani, Generalized Coded Parity Packet Transmission for Gap Value Control over Parallel Channels, submitted to IEEE Communications Letters, June M. K. Gurcan, Hadhrami Ab. Ghani and Irina Ma, The Interferencereduced Energy Loading for Multi-code Coded Packet Transmission, submitted to IEEE Journal on Selected Topics in Communications, June M. K. Gurcan, Irina Ma and Hadhrami Ab. Ghani, The System Value Approach for Interference Cancellation in Multi-channel Systems, submitted to IEEE Transactions on Vehicular Technology, June

6 Published Conference Papers H. Ghani and M. Gurcan, Rate multiplication and two-group resource allocation in multi-code CDMA networks, IEEE 20th International Symposium on Personal, Indoor and Mobile Radio Communications, 2009, pp A. Ghani, M. Gurcan, and Z. He, Two-Group Resource Allocation with Channel Ordering and Interference Cancellation, IEEE Wireless Communications and Networking Conference (WCNC), 2010, pp M. Gurcan and H. Ab Ghani, Small-sized packet error rate reduction using coded parity packet approach, IEEE 21st International Symposium on Personal Indoor and Mobile Radio Communications (PIMRC), 2010, pp Z. He, M. Gurcan, and H. Ghani, Time-efficient resource allocation algorithm over HSDPA in femtocell networks, in IEEE 21st International Symposium on Personal, Indoor and Mobile Radio Communications Workshops (PIMRC Workshops), 2010, pp Accepted Manuscripts for Conference Publication Mustafa Gurcan, Irina Ma, Hadhrami Ab Ghani, and Zhenfeng He, Complexity Reduction for Multi-hop Network End-to-End Delay Minimization, accepted for 26th International Symposium on Computer and Information Sciences, Hadhrami Ab Ghani, Mustafa Gurcan, and Zhenfeng He, Cross-layer Optimization with Two-group Loading for Ad-hoc Networks, accepted for 26th International Symposium on Computer and Information Sciences, M. K. Gurcan, Hadhrami Ab. Ghani and Irina Ma, Successive Interference Cancellation for Two-group Resource Allocation, accepted for IEEE 22nd International Symposium on Personal Indoor and Mobile Radio Communications (PIMRC), Mar

7 Nomenclature General Notation A Constant A Matrix a Vector a Index or variable P (.) Probability of the argument min(.,.) The minimum of the two arguments min x f(x) The minimum value of f(x) for all values of x max(.,.) The maximum of the two arguments max x f(x) The maximum value of f(x) for all values of x (.) T Transpose (.) H Hermitian transpose I K a i,j = [A] i,j E{.} A K K-dimensional identity matrix The entry of matrix A at row i and column j Expectation of the argument diag(a 1, a 2,..., a K ) A diagonal matrix with the scalar input arguments, a 1, a 2,..., a K as the diagonal elements Roman Symbols B E T E s E k J The number of bits in a Turbo-coded parity packet The total energy available per symbol The energy allocated per symbol The energy allocated per symbol at channel k The number of symbols in a Turbo-coded parity packet 7

8 K M N N 0 2 N U N P N p1 N p2 N s,e,l N re N P b V Z b p d min p par m r in r out y k c e d e u v he,l n e r e x e The number of parallel coded-channels employed The number of constellation points in a modulation scheme The spreading factor The single-sided noise power spectral density The length of the source or parity packet The number of the cyclic redundancy check bits The number of the first parity bits produced by a Turbo encoder in a packet The number of the second parity bits produced by a Turbo encoder in a packet The total number of ones in vector h e,l The maximum number of permissible retransmissions The spreading factor of the employed signature sequences The probability of packet error when one external extrinsic LLR is used The number of source packets The maximum number of realizable parity packets using the coded parity packet approach The p-th data rate in bits per symbol realizable by a modulation and coding scheme The minimum distance between two constellation points The index of the data rate chosen using an equal energy loading or equal signal-to-noise ratio loading The number of parity bits generated per information bit The number of channels loaded with a higher data rate when the two-group resource allocation is in place The inner code rate The outer code rate The data rate in bits per symbol for channel k The e-th outer encoded vector The e-th inner encoded vector The v-th information source vector A vector which identifies the l-th unique subset of the extrinsic LLRs The e-th noise vector The e-th received symbol vector The e-th coded symbol vector 8

9 A A e,l ext C C k C out G H ext H base H e J J sel N P P ord Q e R U S b m i The diagonal amplitude matrix for the transmitted symbols The matrix whose columns represent the extrinsic log-likelihood ratio vectors corresponding to the l-th subset for the e-th parity packet The covariance matrix The covariance matrix for channel k when successive interference cancellation is implemented The outer encoded matrix The generator matrix for the coded parity packet approach The extended parity check decoding matrix The base parity check matrix The recovery matrix used in hard decoding the failed packets The upshift/downshift matrix The parity selector matrix to produce G The additive white Gaussian noise matrix The parity generator matrix The channel ordering matrix The receiver signature sequence matrix which includes the intersymbol interference components The received signal matrix The source information data matrix A set of symbols whose the m i -th bit that forms the symbol equals to b, where b 0, 1 9

10 Greek Symbols β Γ Λ() Λ apri () Λ ext () Λ Ext () Λ apos () ε σ 2 ν v Λ ext The bit granularity The gap value The soft channel output of the argument The apriori log-likelihood ratio of the argument The extrinsic log-likelihood ratio of the argument The external extrinsic log-likelihood ratio of the argument The aposteriori log-likelihood ratio of the argument The number of parity packets The white Gaussian noise variance The v-th information source vector appended with a cyclic redundancy check byte The extrinsic log-likelihood ratio matrix whose columns represent all extrinsic log-likelihood vectors Superscript and Subscript e j k l n p v z The parity packet index The symbol index in the transmitted symbol packet The index referring to the k-th channel The external extrinsic log-likelihood ratio vector index The information bit index in the source packet The data rate index The source packet index The index of the decimal values converted to binary to form the parity selector matrix, P 10

11 Acronyms AMC AWGN BER BICM CDMA CPP CRC ER EREE HDPIC HSDPA ISI LC LDPC LLR LTE MAI MCS MIMO MMSE MTF PER PIC QAM QoS SIC SNR SINR UBRT WCDMA Adaptive modulation and coding Additive white Gaussian noise Bit error rate Bit interleaved coded modulation Code division multiple access Coded parity packet Cyclic redundancy check Equal rate margin adaptive loading scheme Equal rate and equal energy loading scheme Hard decision parallel interference cancellation High speed downlink packet access Intersymbol interference Levin-Campello Low density parity check Log-likelihood ratio Long term evolution Multiple access interference Modulation and coding scheme Multiple-input and multiple-output Minimum mean square error Multiple turbo-fountain Packet error rate Parallel interference cancellation Quadrature amplitude modulation Quality of service Successive interference cancellation Signal-to-noise ratio Signal-to-interference and noise ratio Union Bhattacharyya rate threshold Wideband code division multiple access 11

12 Contents 1 Introduction Background Research Problem Description Research Methodology Thesis Outline List of Contributions Coded Parity Packet Introduction Related Work System Model The Encoding Process The Outer Encoding Module The Inner Encoder Modulation The Decoding Process Apriori LLR Generation Aposteriori LLR Generation External Extrinsic LLR Generation Extended Parity Check Decoding Matrix, H ext Associated External Extrinsic LLR Matrices for Encoded Packets, ν e The CPP Hard Decoding Technique Numerical Results i

13 2.7 Conclusion Two-group Resource Allocation Implementation with Coded Parity Packet Approach Introduction Related Work System Model The Chip-matched Filtering Model The Multi-code Transmission Model The Wasted SNR Problem: Equal Rate and Equal Energy Allocation The Iterative Energy Calculation Process Margin Adaptive Optimization Rate Adaptive Optimization Rate Adaptive Optimization Formulation The Two-group Resource Allocation Scheme for Interferencefree Channels Rate Adaptive Optimization Techniques over Channels with ISI and MAI Performance Enhancement for The Two-group Resource Allocation Scheme Channel Ordering Two-group Resource Allocation Scheme: Block Data Transmission Simulation Initializations and Definitions The p and m Calculation The Transmission Process The Despreading Process Numerical Results Implementation of The Two-group and Existing Resource Allocation Schemes with and without ISI The Bit Granularity Factor in Performance of Resource Allocation Schemes ii

14 3.8.3 Channel Ordering Schemes Evaluation Computational Complexity Comparison Conclusion Simplified Loading Schemes with Interference Cancellation and Coded Packet Transmission Introduction Related Work Problem Formulation An SIC-based Two-group Resource Allocation Scheme System Model Channel Ordering Schemes An SIC-based Two-group Rate and Energy Calculation Method The Step-by-step SIC and CPP-based Receiver Implementation Performance Evaluation Total Normalized Mean Square Error Observation Total Average Received SNR and System Throughput Observation The CPP-based Fixed Energy Loading Technique System Model Problem Description Fixed Energy Loading with Adaptive Coded Parity Packet Transmission Numerical Results Conclusion Conclusion and Future Direction Future Work A The Normalized MMSE Despreading Filter Coefficient Vector162 iii

15 B The SIC-based Energy Equation Derivation 163 C The SIC-based Energy Equation Derivation 166 References 168 iv

16 List of Figures 1.1 The gap value reduction using the CPP coding scheme The proposed MCS model The transmitter design with the proposed encoding scheme The summarized algorithm for the proposed CPP encoding scheme Gray mapping for 16-QAM symbols The block diagram of the proposed receiver model The illustration of the external extrinsic LLR generation process The summarized algorithm for the proposed CPP decoding scheme The packet error rates recorded when the proposed scheme is implemented at a packet size of 84 bits for b p = 1.00 bits per symbol, V o = 3, ε o = The packet error rates recorded when the proposed scheme is implemented at a packet size of 84 bits for b p = 1.33 bits per symbol, V o = 2, ε o = The packet error rates recorded when the proposed scheme is implemented at a packet size of 84 bits for b p = 1.50 bits per symbol, V o = 3, ε o = The packet error rates recorded when the proposed scheme is implemented at a packet size of 84 bits for b p = 2.25 bits per symbol, V o = 3, ε o = The packet error rates recorded when the proposed scheme is implemented at a packet size of 256 bits for b p = 1.33 bits per symbol, V o = 2, ε o = The packet error rates recorded when the proposed scheme is implemented at a packet size of 256 bits for b p = 1.5 bits per symbol, V o = 3, ε o = v

17 2.14 The packet error rates recorded when the proposed scheme is implemented at a packet size of 256 bits for b p = 1.6 bits per symbol, V o = 2, ε o = The packet error rates recorded when the proposed scheme is implemented at a packet size of 256 bits for b p = 2.25 bits per symbol, V o = 3, ε o = The effective data rates after retransmissions of 84-bit packets The effective data rates after retransmissions of 256-bit packets The average number of internal Turbo iterations taken to decode a received packet for b p = 1.33 bits per symbol An example of multi-code transmission networks:hsdpa The continuous and discrete time domain model for transmission over a single channel The multi-code transmission system model with matrix and vector representations A graphical illustration of the resource allocation based on the incremental energy model A multi-code transmission system in which the two-group resource allocation scheme is incorporated The total received SNRs achieved at the input of the decoding units when a range of total input SNRs are fed into the twogroup loading (o-tg) scheme and the equal rate margin adaptive loading (ER) scheme with discrete data rates of zero gap values The system throughputs realizable when a range of total input SNRs were provided for the two group (o-tg), the equal rate margin adaptive loading (ER) and the equal energy and equal rate loading (EREE) schemes The total average received SNRs achieved at the input of the decoding units when a range of total input SNRs were fed into the two-group loading (o-tg) scheme and the equal rate margin adaptive loading (ER) scheme The total average received SNRs achieved at the input of the decoding units when a range of total input SNRs were fed into the two-group loading (o-tg) scheme with large bit granularities, along with the equal rate and equal SNR allocation (ER) scheme 103 vi

18 3.10 The system throughputs realizable when a range of total input SNRs were provided to the two group (o-tg), the equal rate and equal SNR (ER) and the equal rate and equal energy (EREE) schemes at zero gap values The system throughputs realizable when a range of total input SNRs were provided to the two group (o-tg), the equal rate and equal SNR (ER) and the equal rate and equal energy (EREE) schemes The system throughputs realizable when a range of total input SNRs were provided for the two group (o-tg) and the equal rate margin adaptive loading (ER) schemes using the single and double-loaded channel ordering schemes The system model The Receiver Structure The total MSE obtained when the two-group resource allocation scheme and the equal rate margin adaptive loading scheme were implemented with SIC The total average received SNR obtained when the two-group resource allocation scheme was implemented with and without SIC The system throughput obtained when the SIC-based two-group (o-tg SIC) loading was implemented The system throughput obtained when the SIC-based two-group (o-tg SIC) loading was implemented The multi-code transmission system block diagram The packet error rates for the proposed CPP scheme at b p = 1.33 bits per symbol, V o = 2, ε o = 3, and α = 4 at varying SNRs in the channels with different means and variances The packet error rates for the proposed CPP scheme at b p = 1.78 bits per symbol, V o = 2, ε o = 3, and α = 4 at varying SNRs in the channels with different means and variances The packet error rates for the proposed CPP scheme at b p = 1.33 bits per symbol, V o = 2, ε o = 3, with varying adjustment parameter values α = 1, 2, 3, 6. The packet error rates for the Turbo-QAM at the same data rate are also plotted vii

19 4.11 The packet error rates for the proposed CPP scheme at b p = 1.6 bits per symbol, V o = 2, ε o = 3, with varying adjustment parameter values α = 1, 3. The packet error rates for the Turbo- BICM QAM at the same data rate are also plotted viii

20 List of Tables 2.1 The adjustment of the number of the subsets of the parity packets The parameters used for the simulation Comparison of the number of matrix inversions The number of matrix inversions and realizable discrete total data rates for P = 14, P small = 6, K = 15 and I max = The parameters used for the simulation Comparison of the number of matrix inversions and realizable discrete rates ix

21 CHAPTER 1 Introduction 1.1 Background High speed downlink packet access (HSDPA) [1], which is implemented over the wideband code division multiple access (WCDMA) networks, is one of the most widely adopted high speed broadband standards. Various features are provided to enhance the system throughput including adaptive and modulation coding (AMC) and multi-code transmission. To increase the practical achievable system throughput in order to improve performance of HSDPA systems is one of the main focal areas in the research community. It has been indicated in [2] that there is still a substantial room for improvement to bring the practical achievable system throughput closer to the theoretical upper bound. With the incorporation of additional features such as multiple antennas at both the transmitter and the receiver ends [1], any proposed designs to further improve performance of HSDPA systems should be computationally efficient whilst keeping the achievable system throughput relatively high. In this thesis, an integrated multi-code transmission system model, which incorporates a capacityapproaching channel coding scheme, an efficient resource allocation scheme and a successive interference cancellation scheme, is designed to address these challenges. In the next section, these main challenges as well as other related 1

22 CHAPTER 1. INTRODUCTION research questions addressed in this thesis are described. 1.2 Research Problem Description Consider a multi-code HSDPA communication system incorporated with a modulation and coding scheme (MCS) configurable to operate at a set of realizable data rates. For each of the rates, this research project attempts to design a modulation and coding method such that the gap value [3], which is the ratio between the required signal-to-noise ratio (SNR) and the minimum theoretical SNR to hit a target packet error rate, can be controlled by adjusting the number of alternative additional input information values for the decoding process without changing the original rate. This is realized by generating an adjustable number of source and parity packets at the transmitter end without changing the rate. The external extrinsic log-likelihood ratios, which are the additional input information values, are alternately fed into the decoding process until the received packet is successfully decoded or all external extrinsic log-likelihood ratios have been tested or fed into the decoder. For brevity, this MCS is called the coded parity packet (CPP) scheme. Reduced gap values yield lower energy requirement to hit a low target packet error rate, which is essential to reduce the number of retransmissions, hence the end-to-end delay, as well as the energy consumption in wireless ad-hoc networks. An example of a reduced gap value is shown in Figure 1.2. The gap value is reduced by more than 2 db using the CPP coding scheme configured at 1.5 bits per symbol as opposed to the Turbo coding scheme configured at 1.33 bits per symbol, where both coding schemes were run with 16-QAM scheme. The gap value is reduced by increasing the adjustment parameter, α, which increases the number of alternative additional input information values for the iterative decoding process at the receiver end. Further explanation on this adjustment parameter and its effects on improving the decoding process and reducing the gap value is given on page 19 and page Page 2

23 CHAPTER 1. INTRODUCTION 25(The iterative decoding process) Packet size = 256 bits, Generator polynomial, [23,35] 8 α = 1 α = 2 α = Turbo 10 2 PER Es in db N0 Figure 1.1: The gap value reduction using the CPP coding scheme This proposed coding technique is integrated with a fixed signal-to-noise ratio (SNR) loading scheme which divides transmitting channels into two groups, each loaded with a distinct fixed SNR. As it loads two adjacent data rates to the employed code channels, this loading scheme is referred to as a twogroup resource allocation scheme. With two groups of channels loaded with different fixed SNRs to realize two distinct data rates, the total transmission data rate is increased since a larger amount of energy is utilized to realize these data rates. When a set of realizable discrete data rates with relatively large bit granularities, which are the differences between two adjacent data rates, is provided, the two-group resource allocation scheme has been observed to produce an approximately equal system throughput as compared to that of the existing loading schemes, which are provided with a set of discrete data rates of large bit granularities. The two-group resource allocation scheme is further improved by eliminating Page 3

24 CHAPTER 1. INTRODUCTION any inverse matrix operations by employing a successive interference cancellation (SIC)-based energy calculation method. With this SIC-based energy calculation approach, the aforementioned CPP scheme can be applied to provide symbols including the past and the future symbols for the interference cancellation process. Eliminating matrix inversions is essential for practical implementation in order to further enhance performance of multi-code HS- DPA systems. This computationally efficient SIC-based two-group resource allocation scheme may also be considered to be implemented in the multipleinput multiple-output (MIMO) HSDPA systems [1], which is more complicated than the existing single-input single-output (SISO) multi-code HSDPA systems. Apart from allocating two adjacent data rates over two groups of code channels, the total energy may also be equally distributed over each channel while loading an equal data rate to each channel corresponding to the average SNR of all channels. In order to control the gap value, the CPP scheme is applied to realize this equal data rate on each channel. Apart from controlling the gap value, this equal energy loading method is also computationally efficient as it requires no energy calculation. In the next section, the research methodology adopted to solve these research problems is briefly described. 1.3 Research Methodology The key steps applied to solve the aforementioned research problems are summarized as follows: Research Focus With the problems described in the previous section, the main focus of this research project is the reduction in the gap value achieved in the downlinks of Page 4

25 CHAPTER 1. INTRODUCTION multi-code HSDPA communication systems. The reduced gap value is essential in order to minimize the required energy to transmit the data, hence enhance the total transmission data rate for a given total constrained energy. Assumptions In Chapter 2, a single-link white Gaussian channel is assumed for the implementatio of the proposed channel coding scheme. In Chapter 3, it is assumed that the channel side information including the channel impulse response and the noise variance is known to both the transmitter and the receiver. The noise is assumed to be the additive white Gaussian noise. To demonstrate the feasibility of implementing the proposed system design in practical systems, ISI and MAI components, which severely degrade the quality of the received signal, are also considered. As the channel side information might not be perfectly known in practice, the capacity-approaching CPP scheme is implemented with the proposed two-group resource allocation scheme in Chapter 3 and 4 to ensure that the received data are successfully decoded and detected although the received SNRs vary due to the imperfect knowledge of the channel side information, including the received SNR, and the noise variance. An alternative fixed energy loading scheme with the CPP method is also implemented in Chapter 4 when no channel side information is known. All code channels in the multi-code HSDPA systems are dedicated to a single user as the focus of this thesis is the gap value reduction and not the resource scheduling. However, the design developed in this thesis may be extended for solving other research problems including the resource scheduling. Solution Methods Knowing the research problems, solution methods are developed based on the assumptions made. The solution methods include the CPP coding scheme, the Page 5

26 CHAPTER 1. INTRODUCTION two-group resource allocation method and the SIC-based loading technique. In multi-code HSDPA systems, reducing the gap values and consumed energies as well as improving the total transmission data rates using these solution methods should consider the presence of both ISI and MAI components, which have been assumed earlier, in order to render the proposed design feasible in practice. Besides improving performance, the required computational complexity should also be monitored. The solution methods described in this thesis are also compared to the existing methods in solving the research problems addressed. Results Production and Verification The solution methods proposed in this thesis have been continually tested by different people under distinct assumptions to produce and verify the desired results. Furthermore, these results have also been compared to those of the existing and latest schemes. By doing multiple tests under different assumptions, the application of the proposed solution methods may be generalized to a wider range of areas in practice especially the improvement in the total transmission data rate over the downlinks of the multi-code HSDPA systems in the presence of ISI and MAI. Revision Process Apart from repetitively testing the proposed solution methods or designs, conference and journal publications have significantly contributed towards improving and rectifying the problems encountered in designing the proposed solution methods. The comments received during the paper revision process have given useful feedbacks to improve and verify the solutions. Page 6

27 CHAPTER 1. INTRODUCTION 1.4 Thesis Outline In short, this thesis consists of the following chapters 1. Chapter 1: The first chapter describes the background, the research problems and the research methodolgy taken to complete the thesis. 2. Chapter 2: The second chapter gives an account of the CPP scheme, which reduces the gap values without changing the data rate. This reduction is further applied for improving the two-group resource allocation scheme described in Chapter Chapter 3: The third chapter presents a two-group resource allocation scheme which is integrated with the CPP scheme to minimize the required energy to load the data rates, and hence increase the total transmission data rate. 4. Chapter 4: This chapter enhances the two-group loading method in the previous chapter by further reducing the required energy when successive interference cancellation is implemented to load the data rates whilst keeping the computational complexity low. 5. Chapter 5: The last chapter concludes the thesis and highlights a number of future directions to further improve and apply the solution methods developed in the thesis. 1.5 List of Contributions The contributions made by completing this research project are listed as follows 1. A capacity-approaching coding scheme referred to as CPP coding has been developed, which is able to reduce the gap value achieved using Page 7

28 CHAPTER 1. INTRODUCTION the existing Turbo codes by more than 2dB when small packet sizes of around 100 bits are transmitted at data rates between 1 to 2 bits per symbol. (Chapter 2) 2. A two-group resource allocation scheme, which is implemented with large bit granularities, that produces an approximately equal system throughput to that of the existing equal rate margin adaptive loading scheme, which is implemented with small bit granularities. (Chapter 3) 3. Improvement of approximately 5 db achieved by the two-group resource allocation scheme as compared to that of the equal rate and equal energy loading scheme. (Chapter 3) 4. Reduction in the amount of energy to realize the target total data rate using an SIC-based two-group resource allocation scheme is used. (Chapter 4) 5. Reduction in the computational complexity using the SIC-based energy calculation method, which requires no matrix inversions. (Chapter 4) 6. Reduction in the gap value when a fixed energy loading is incorporated into the CPP coding scheme for multi-code systems. When compared with existing schemes, the achieved gap value is observed to be lower by up to 2 db. (Chapter 4) The final aim of this thesis is to design a scalable and computationally efficient multi-code transmission system model, which is proposed as an alternative to further improve HSDPA systems. Since the current practical achievable system throughput is still lower than the theoretical upper bound, an integrated transmission system which combines the proposed packet-based channel coding scheme with the two-group resource allocation technique and the SIC scheme has been developed in this thesis as a possible solution which might be considered to be incorporated in the future version of the multi-code HSDPA systems. Page 8

29 CHAPTER 1. INTRODUCTION Realizing that the number of matrix inversions required is a serious issue in the resource allocation as well as the signal detection process, the proposed integrated system, which employs the SIC-based energy calculation technique, eliminates these matrix inversions whilst maintaining and improving the achievable system throughput. This computationally-efficient system may also be considered as a solution for improving the latest broadband technologies such as MIMO HSDPA and long term evolution (LTE) systems. Page 9

30 CHAPTER 2 Coded Parity Packet 2.1 Introduction The increasing demand for high-speed wireless broadband technologies stimulates much interest in both the research community as well as industry. Various communication models and products have been developed and proposed to improve system performance for serving this growing need. One of the key measures to gauge system performance is the gap value [3], which is defined as the difference in db between the theoretical signal-to-noise ratio (SNR) achievable from Shannon s equation [4] for a given data rate in bits per symbol and the experimental SNR required by a system under consideration in order to achieve the same data rate at a target error rate. For reducing the gap value at a given data rate, the transmission system model must be designed with the capability of successfully transmitting and receiving data at the particular data rate for a given target error rate. This must be achieved by using as low an SNR as possible, which is lower bounded by the theoretical SNR calculated from Shannon s equation. One of the commonly adopted methods used to improve performance of the transmission system model by reducing the gap value is channel coding. Various 10

31 CHAPTER 2. CODED PARITY PACKET channel coding techniques are proposed in literature, which mainly classified as block and convolutional codes [5]. The block codes are produced from linear combinations of the information sequences, whereas the convolutional codes are generated from the sequences of codes in both the current and the previous information sequences. However, it has been observed that most of the existing codes perform well at certain range of code rates. Turbo codes [6], which are commonly regarded as capacity-approaching convolutional codes, have been demonstrated to reduce gap values at low code rates although they suffer from residual error floor problems. At high code rates, block codes such as low density parity check (LDPC) codes [7] are reported to perform better whilst reducing computational costs. To reap the benefits of both block and convolutional codes, concatenated codes are also designed in literature, such as in [8]. Reduced gap values at various data rates are essential to ensure that low target packet error rates are achievable with relatively low required energies. In wireless ad-hoc networks, a low target packet error rate is important to reduce retransmissions, which will shorten the end-to-end delay. Furthermore, the low energy requirement renders lower energy consumption, which can also be reduced as the number of retransmissions decreases. In this chapter, a concatenated channel coding scheme which combines a simple block coding technique with a standard Turbo coding scheme is developed to significantly reduce the gap values, rendering lower energy requirements for packet transmission in wireless ad-hoc networks. 2.2 Related Work Some of the most commonly studied and implemented capacity-approaching coding schemes which are Turbo codes [6], LDPC codes [7] and repeat accumulate (RA) codes [9]. Most of these codes are chosen as the standard forward error correction (FEC) techniques and incorporated into modulation schemes Page 11

32 CHAPTER 2. CODED PARITY PACKET such as quadrature amplitude modulation (QAM) schemes in current wireless communication systems including the wireless broadband code division multiple access (CDMA) systems. In this chapter, channel coding schemes especially concatenated coding techniques [8] based on Turbo codes are studied in order to improve coding performance and reduce the gap value. Concatenating Turbo codes with other channel codes is mainly motivated by the need for eliminating the error floor problem experienced by the standard Turbo coding schemes at considerably high SNRs. Since the focus of this chapter is on Turbo-based coding schemes, less emphasis is given on other channel coding schemes such as LDPC and RA codes. Turbo codes have been extensively studied in literature, yielding numerous variants of modulation and coding schemes (MCSs) developed and proposed to improve system performance. A commonly adopted design of Turbo-based MCSs was developed by concatenating a Turbo coding scheme with a modulation scheme, such as in [10], for instance. These MCSs can be further upgraded by incorporating interleavers to avoid burst errors on the received bits, as presented in a system known as the bit-interleaved coded modulation (BICM) scheme [11 15]. Alternatively, Turbo codes can also be combined with modulation schemes, as implemented by Ungerboeck [16, 17], which are further modified and enhanced in [18 21]. In addition to the aforementioned designs, the applications of Turbo codes are also extended in developing multi-level coding schemes [22, 23] such as in [24]. Turbo codes are reported to produce low gap values especially at low code rates, but suffer from a residual error floor problem when the SNR is considerably high. Using a random puncturing method, a Turbo-based design known as variablerate Turbo-BICM scheme was developed in [14]. The corresponding achievable capacity by this scheme was modelled using a Union Bhattacharyya Rate Page 12

33 CHAPTER 2. CODED PARITY PACKET Threshold (UBRT) equation, which was generically produced from [3] based on the Shannon s equation [4]. The concept of message passing or belief propagation, as commonly employed in LDPC codes, was implemented in [25] to improve the corresponding iterative soft decoding performance on the receiver end in order to reduce the gap value. The messages or the extrinsic log-likelihood ratios passed during the decoding process are generated from every received parity bits as well as the received systematic bit on the receiver end. Although satisfactory performance is reported, this so-called Multiple Turbo-Fountain (MTF) coding scheme is limited to low code rates, in the form of 1/(1 + par), where par denotes the number of parity bits generated along with an information bit during the encoding operation. Furthermore, the problem of residual error floor commonly observed in Turbo codes is also inherent in this coding scheme. A possible alternative of realizing multiple distinct discrete data rates is by concatenating different codes [8], which generally produces longer codes. The simplest form of this type of coding is a concatenation of two codes known as outer and inner codes. Capacity-approaching codes such as Turbo codes can be used as one of the codes being concatenated. Turbo-Hadamard codes developed in [26 28] for example, were implemented by concatenating Hadamard and Turbo codes. This scheme, although achieving low gap values, generates very long codes, which requires high computational loads. Another variant of Turbobased concatenated codes is proposed in [29], which also achieves low gap values. However, the code rates realized are also limited and the required computational cost to perform the iterative decoding process is high. In this chapter, a new method of realizing multiple distinct data rates at relatively low gap values is proposed. An efficient parity packet generation approach is developed to produce an adjustable number of source and parity packets, each of which is further inner encoded using a standard Turbo encoder. With- Page 13

34 CHAPTER 2. CODED PARITY PACKET out changing the code rate, the number of these packets can be adjusted using an adjustment parameter α to control the number L ε = 2 α(εo Vo) 1 of subsets of the associated packets, given the initial number of source and parity packets as V o and ε o V o respectively. These subsets of associated packets, which are received on the receiver end, are used to produce a number of alternative additional input information values known as external extrinsic log likelihood ratios (LLRs) for the iterative decoding process. Furthermore, a simple hard decoding technique can also be invoked based on these subsets to recover the failed source packets. Tests run at small-sized packets using this proposed coding scheme demonstrate that significant gap value reduction is achieved. This reduction is highly applicable to reduce energy consumption while keeping the packet error rate low [30] to reduce the number of retransmissions and end-toend delay in wireless ad-hoc networks. The rest of this chapter is organized as follows: the next section describes the system model in which the proposed MCS is implemented. A detailed account of the encoding scheme developed is given in Section System Model Consider a model of a single-link communication system implemented between two interacting nodes in a wireless ad-hoc network, where a packet-based concatenated coding scheme comprising an outer and inner coding scheme is incorporated, as appears in Figure 2.1. The information sequences intended for transmission are generated in blocks of data represented as a (K 1)-dimensional vector u, each element of which is denoted as u k {±1} for k = 1,..., K. Each of the data blocks is further segmented into ((N U N P ) 1)-dimensional vectors u v for v = 1,..., V. Then, an outer encoder is used to encode these packets to yield ε packets, which are denoted as ((N U N P ) 1)-dimensional vectors c e, before being appended with cyclic redundancy check (CRC) bytes, Page 14

35 CHAPTER 2. CODED PARITY PACKET each of which having N P bits, to yield (N U 1)-dimensional vectors, ν e. These CRC-appended packets are then inner-encoded to produce (B 1)-dimensional vectors d e. Each of these so-called coded parity packets is modulated to generate (J 1)-dimensional packets of x e, which are transmitted over an additive white Gaussian noise channel, rendering a noise-corrupted received packet of r e for e = 1,..., ε. On the receiver end, the received packet, r e is first demodulated before being passed on to the decoder. Since a soft-in hard-out decoding process is used, a (B 1)-dimensional channel soft output vector, Λ( d e ), needs to be determined from each received packet r e to be fed into the decoding process. The channel soft output is a type of log-likelihood ratio, representing the reliability of the received packet. This soft information is fed into the soft-in-hard-out iterative soft decoding process to perform the actual decoding operation, from which the estimate of the (N U 1)-dimensional cyclic redundancy check (CRC)-appended parity packet ˆ ν e is made. Hard decoding is carried out based on the succesful packets to recover the failed packets, if any. Binary source u u v c e ν e de Segmentation Outer encoding Cyclic redundancy check Inner encoding Modulation x e n e r e = x e + n e Binary sink ˆ u Desegmentation ˆ u v Hard decoding ˆ c e Cyclic redundancy check ˆ ν e Iterative soft decoding Λ( d e ) Demodulation Figure 2.1: The proposed MCS model As highlighted in Figure 2.1, the focus of this chapter is to describe the encoding and decoding designs developed with the aim of achieving reduced gap values. In the next section, the design of the proposed encoding process is presented. Page 15

36 CHAPTER 2. CODED PARITY PACKET 2.4 The Encoding Process The methods presented in this section and most of the following sections are based on a manuscript presented in [31], which has been submitted to European Transactions on Telecommunications and refined from the original document submitted to IEEE Transactions on Communications. As can be seen in Figure 2.1, the proposed encoding scheme consists of two components; the outer encoding module and the inner encoding module. The outer encoding module is a packet-based encoding scheme which generates parity packets by combining the source packets, as will be described in the following section. In this section, a technique of selecting and generating ε V parity packets from V source packets is developed such that 2 ε V 1 distinct subsets of associated packets can be generated to produce L ε = 2 ε V 1 external extrinsic LLRs for each source packet received at the receiver end. Then, an adjustment parameter α is introduced such that the number of subsets is controlled to be L ε = 2 α(εo Vo) 1, where V o and ε o V o are the initial number of source and parity packets respectively and V = αv o, ε = αε o. As the number L ε of external extrinsic LLRs increases, the probability of successfully decoding a received packet also increases as each of these LLRs can be alternately fed into the decoding process to recover the received packet. Figure 2.2 shows the design of the transmitter, developed on the basis of the proposed encoding scheme. The account of this encoding scheme, which is indicated inside a closed dotted line in the figure, will be presented in the next subsection The Outer Encoding Module The proposed outer encoder is designed to generate ε packets [ c 1,..., c e,..., c ε ] = C out, each of which with a dimension of ((N U N P ) 1), from V source pack- Page 16

37 CHAPTER 2. CODED PARITY PACKET c 1 CRC ν 1 Channel encoder d1 Modulator x 1 u Segmentation u 1 u 2 u ε Outer encoder Cout = U G c 2 c ε CRC ν 2 Channel encoder d2 Modulator x 2 Parallel-to-serial conversion to the channel CRC ν ε Channel encoder dε Modulator x ε Figure 2.2: The transmitter design with the proposed encoding scheme ets [ c 1,..., c v,..., c V ] as follows C out = U G, (2.1) where U = [ u 1,..., u v,..., u V ] is an ((N U N P ) V )-dimensional matrix representing the block of input information data and is the element-by-element vector or matrix multiplication and bit-wise modulo-2 addition operation. The (V ε)-dimensional matrix G is the CPP generator matrix used to produce ε V desired parity packets by bit-wise XOR combining two or more of the V source packets, in addition to the existing V source packets. When a parity packet is generated from d source packets, this parity packet is referred to as a packet of degree d. In simpler terms, the degree d of a particular packet signifies the number of packets called neighbours which form the packet through a bit-wise XOR combination between all of them. With V source packets, there are a maximum of 2 V V 1 parity packets which can be generated through this encoding technique, in addition to the source packets. Therefore, a generator matrix which can produce a maximum of 2 V 1 parity packets including the source packets can be defined as [ G (m) = I V P ], (2.2) Page 17